LST determination of different urban growth patterns: A modeling procedure to identify the dominant spatial metrics

Abstract

The growing threat of urban heating has attracted considerable scholarly attention to their modeling and prediction. To improve the urban land surface temperature (LST) determination, this study assumed that the spatial pattern of built-up patches matters in LST prediction. To test this idea in a rapidly-urbanizing landscape, different urban growth types including infilling, edge expansion and outlying patterns and their LST layers were extracted from Landsat images in 1992, 2002, 2012, and 2022. The mean LST of each growth pattern was then modeled using the multiple linear regression (MLR) analysis and an array of independent variables related to the spatial formation and LST of the newly-grown and previously-existed urban patches. Results of the best infilling MLR model (R2 = 0.579) showed that the highest mean LST of the infilling growth is expected to be around the center of large focal patches. The mean LST of the edge expansion growth patches were found to be influenced by the mean LST of the edge of their focal patches and their shape structure (R2 = 0.362). The area of the outlying growth patches was also selected as the sole predictor of their mean LST (R2 = 0.334). According to the findings, the mean LST of the outlying, infilling and edge expansion growth patterns are influenced by the landscape composition, configuration and structure, respectively. Our results corroborate the idea that the performance of the urban LST prediction and their practical implications can be improved when the built-up landscape is divided into more spatially similar growth classes.